Scheduling of scientific workflows in the ASKALON grid environment

TLDR

Scheduling is a key concern for the execution of performance-driven Grid applications. The study compares existing scheduling approaches for scientific workflow applications in a Grid environment. The authors evaluate genetic, HEFT, and myopic algorithms, comparing incremental partitioning to full-graph scheduling on real-world balanced and unbalanced scientific workflows. Full-graph scheduling with HEFT outperformed the other strategies.

Abstract

Scheduling is a key concern for the execution of performance-driven Grid applications. In this paper we comparatively examine different existing approaches for scheduling of scientific workflow applications in a Grid environment. We evaluate three algorithms namely genetic, HEFT, and simple "myopic" and compare incremental workflow partitioning against the full-graph scheduling strategy. We demonstrate experiments using real-world scientific applications covering both balanced (symmetric) and unbalanced (asymmetric) workflows. Our results demonstrate that full-graph scheduling with the HEFT algorithm performs best compared to the other strategies examined in this paper.